Abstract
Patients in forensic mental health care have a difficult journey through inpatient rehabilitation and re-integration into the community. Risk assessment guides this progress, usually with clinician-based processes that use structured risk-assessment tools. Patients' understanding of their own risk is important to inform risk assessment and the chances of successful rehabilitation. The emergence of shared decision-making approaches provides an opportunity to consider shared risk assessment and formulation. We reviewed the literature to explore models of patients' involvement in risk assessment and the impact on outcomes in forensic mental health care. We conducted searches of three databases (Medline, PsychINFO, and EMBASE) to identify papers that employed shared risk understanding for violence risk. Additional records were identified through review of citations, with articles being selected using a predetermined set of inclusion and exclusion criteria. We found five studies that met the inclusion criteria for patient involvement in risk assessment with measurement of construct or predictive validity. The studies employed diverse methodologies, but they suggest that patient involvement in assessing risk is feasible when correlated with staff ratings. There is encouraging evidence of the predictive validity of self-rated risk alongside staff-rated risk assessment.
Shared decision-making (SDM) is a tool proposed in recent years to enhance service-user involvement in mental health care decisions, although this movement has often lacked careful evaluation.1 Tilley et al.2 described user involvement as the extent to which the patient is involved in defining problems and setting the targets that constitute the plan of care. In mental health, SDM is considered a middle ground between paternalism and autonomy, and a potential alternative to coercive interventions.3 A recent meta-analysis found equivocal results for the impact of SDM on outcomes for persons with psychosis, a small effect enhancement of subjective sense of empowerment, and a trend toward a reduction in coercive interventions with SDM.4
In forensic mental health care, a recovery framework must balance a patient's best interests with public safety concerns.5 Forensic recovery presents dual issues of enhancing autonomy while also giving due consideration for public safety. Care is provided within a coercive framework. A recent meta-synthesis of forensic patient accounts of recovery called for developments to enhance patient inclusion to increase patients' sense of safety and understanding in the process of forensic recovery.6 Similar themes have been found in the more recent work of Livingston7 in patients and family members defining success in forensic recovery. It is possible that important information that could guide risk-management plans may be overlooked if patients are not involved in the process. Risk assessment plays a key role in guiding the decision-making processes of risk management.
SDM brings together patients and treatment teams in making health care decisions. Being involved may enhance patients' sense of self-efficacy and responsibility by contributing to important decisions regarding their care. In addition, it gives clinicians a chance to get a better view of the patients' insight into the issues pertaining to their risks. Applied to risk assessment, this involves a joint process of contributing to an understanding of key risk issues and effective risk management.
Given the above, we sought to explore examples of SDM as applied to risk assessment and management of violence in forensic psychiatry. To do so, we conducted a literature review of quantitative studies of shared risk formulation using structured risk-assessment tools in forensic mental health practice. The aims of the review were to describe the methodologies employed and the settings where they were applied, and to examine their reliability and validity.
Methods
Eligibility Criteria
Original studies, systematic reviews, and meta-analyses focusing on shared risk assessment, formulation, or decision-making in forensic psychiatric or correctional settings for mentally disordered offenders with problems of violence were considered for this study. The studies were all in English and had a published status. Studies were identified by searching electronic databases and by scanning reference lists of relevant articles. The search was conducted in the following databases: Ovid MEDLINE 1946 to present, Ovid MEDLINE In-Process and Other Non-Indexed Citations, PsycINFO 1806 to present, Embase Classic + Emblem 1947 to present.
Search
The following search terms were used in all databases: (forensic or prisons or offend), (shared or collaborate) and (risk assessment or risk formulation or treatment plans or care plans). Search strategies were agreed upon by both authors. A total of 245 articles were identified in the search. The retrieved records were screened by reviewing the title and the abstract, and 10 papers were fully reviewed by both authors for final inclusion. Each stage of the selection process was carried out by both investigators. Eligibility assessment was performed independently in an unblended standardized manner by both investigators.
Results
As shown in Fig. 1, five papers met the inclusion criteria (Table 1). Four of the five studies were carried out in forensic inpatient settings,8,–,11 including maximum-, medium-, and minimum-security levels of rehabilitating forensic patients, and two were separate analyses of the same study in a forensic outpatient service.12,13
Summary of study selection and exclusion (flow diagram).
Review of Studies on Shared Risk Formulation in Forensic Population
Settings
Of the inpatient settings, Bjorkly's9 case report was carried out in a well-staffed medium-security forensic psychiatry unit which receives patients with a history of violence for long-term treatment. Fluttert's8 work was carried out in a maximum-security forensic psychiatry hospital with 189 male patients. The remaining settings were integrated forensic hospitals in Australia10 and Ireland.11 The study from The Netherlands12 included three outpatient forensic psychiatry clinics with 632 patients and 58 case managers, with 310 subjects included in the study. The second report from this study is a subset analysis of 196 of these subjects.12 The patients in the Dutch study differed from the other studies in that they were predominantly diagnosed with personality disorders, with only eight percent having a psychotic disorder.
Shared Risk Assessment and Effectiveness
Bjorkly9 presented a case illustration of a risk management approach based on progression ladders, called ProLad. It is a structured contingency management approach for preventing relapse of violence, and it is tailored to individual needs, which require patients to be actively involved in making plans for progress in conjunction with their nurses. The starting point is to establish a therapeutic relationship and involve patients in the planning as soon as possible. The scope is to work on several goals simultaneously to address risk issues and to focus on personal growth. In the case study, the subject was able to identify the early signs of auditory hallucinations and worked in collaboration with nurses to intervene at the initial stages of relapse. The author presented this case as an example of the feasibility of shared risk assessment.
Fluttert et al.8 developed and evaluated a tool called the Early Recognition Method (ERM). The ERM is a guided process in which staff and patients develop a shared understanding of early signs of aggression and implement plans to reduce violence. The ERM was implemented in four phases. In the first phase, the intervention was explained to the patient. Second, a list of early signs of aggression was prepared by the patient with help of the nurse mentor. In the third phase, the patient and staff mentor together monitored patient's behavior to detect early signs of aggression. In the fourth phase, preventive actions were listed in the early detection plan and were implemented to help the patient de-escalate and regulate their behavior. Each session lasted for 30 minutes and was made a part of existing weekly evaluations. The subjects received their usual treatment and the early recognition method in a phased manner where the subjects acted as their own controls in a crossover design. Of the 189 patients eligible for the study, 168 participated; refusers had higher psychopathy scores and more substance-abuse diagnoses. The outcome measures were the number of seclusion events and the severity of inpatient incidents of aggression as rated on the Staff Observation Aggression Scale–Revised (SOAS-R). The study showed that implementation of ERM led to a decrease in the frequency of seclusion events from 219 events in the control phase to 104 in the intervention phase [χ2 = 22.82, p < .001], and the rate of seclusion events per patient per month reduced from a mean of 0.13 to 0.05 (p < .001). There was also a reduction in SOAS-R severity of incidents from 1.38 to 0.50 (p < .001).
Abou-Sinna and Luebbers10 investigated whether staff and patient co-ratings of the Camberwell Assessment of Needs–Forensic (CANFOR-S) tool were related to traditional staff-only rated measures of risk and need as measured with the Health of Nations Outcome Scale–Secure (HoNOS-S) and the Historical Clinical Risk-20 (HCR-20). The setting was an Australian forensic hospital with 72 forensic patients. The authors reported that the CANFOR-S completed jointly by patients and their primary nurses were positively correlated with the clinical and security scales of the HoNOS-S as well as the clinical and risk management scales of the HCR-20.10 The authors suggested that shared risk assessment processes were feasible and that patients could be successfully engaged in identifying risks and needs.10
Davoren et al.11 performed a prospective naturalistic cohort study employing the DUNDRUM Toolkit, a structured professional judgment measure of rehabilitative progress and risks. They employed staff and patient versions of the same tools that were designed to measure progress across program participation (DUNDRUM 3) and risks (DUNDRUM 4). The self-rated versions were created with the involvement of service users. The DUNDRUM 3 and DUNDRUM 4 tools were rated independently by staff and patients, with the patients' self-ratings being withheld from the clinical teams and review tribunals who made decisions about progress during the follow-up period. Higher scores on the DUNDRUM scales indicated higher risk and less successful rehabilitative progress. The authors' hypothesis was that increased concordance between staff and patient ratings of risks and needs would predict clinical progress and conditional discharge over the 14-month follow-up period.11
Only male patients were enrolled in the study; the response rate was 66 percent, and the final sample size was 58 subjects.11 Results showed that the patients' self-ratings on the DUNDRUM-3 and DUNDRUM-4 were overall significantly lower than staff ratings; that is, patients considered themselves to be further along the road to recovery than clinicians did. Cross-correlation showed highly significant correlations between clinician-rated DUNDRUM 3 and self-rated DUNDRUM 3 scores (Spearman correlation coefficient r = .566, p < .001) and between the clinician-rated DUNDRUM 4 recovery scale and the self-rated DUNDRUM 4 recovery scale (r = .712, p < .001).11 Patients moving from higher to lower levels of security during the follow-up period had lower (better) scores on the clinician-rated DUNDRUM 3 program completion scale; lower clinician-rated DUNDRUM 4 scores predicted the same movement, whereas the self-rated scales did not predict the positive movement. Clinician ratings predicted the negative movements, but the patient ratings did not predict those movements, nor did the self-rated scores predict movement between levels of security or conditional discharge.11 However, an interesting finding was that concordance between clinician and patient ratings increased during progression from high- to medium- to low-security settings, and greater concordance meant being closer to discharge. The authors considered that concordance between clinician and patient ratings could be a useful index of the degree to which understanding of risks has become shared.11
The final study, the Risk Assessment and Care Evaluation (RACE) study,12,13 has been reported in two ways, as the results of a randomized control study and as a predictive study within the intervention arm only. The study was a cluster randomized controlled trial of staff- and patient-completed versions of the Short-Term Assessment of Risk and Treatability (START) tool among patients primarily diagnosed with personality disorders compared with treatment as usual.12 The START is a risk assessment tool consisting of 20 items scored as risk factors and strengths. The research team developed a client version of the START called the Client Self-Appraisal (CSA). In the intervention arm, clinicians and patients each completed their own assessments of the START and CSA and then discussed the patient's key strengths and critical vulnerabilities to inform care planning for each year. The control group used neither the START nor the CSA, providing treatment as usual. The primary outcome was recidivism at six-month follow-up. There were 310 subjects in the study, of whom 201 received the intervention (64.8%) and had care plans developed according to the RACE protocol; 72 subjects (23.2%) received the intervention more than once (two to four times).12 Interestingly, there was no significant effect on violent recidivism between the control and intervention groups (odds ratio = 1.46, 95% CI 0.89–2.44, p = .15), which suggests that neither the START nor the CSA utilized in this way contributed to more effective community care plans.12
Van den Brink et al.13 conducted a second analysis of the 201 subjects in the RACE study who completed the full START and CSA assessments to explore the concurrent and predictive validity of staff- and patient-rated measures. A total of 196 subjects (98%) were able to complete the CSA, suggesting it is a feasible instrument. Concordance of client and staff assessments was good for critical vulnerabilities and key strengths, indicating general alignment between staff and patients about the most important issues. Concordance at an individual level for key risk and predictive factors was limited, however, with staff generally recording higher risks and vulnerabilities than patients. Predictive modeling of the START and CSA results for violent and criminal recidivism was considered. Client self-assessments of critical vulnerabilities and key strengths were significant univariate predictors of recidivism and indeed were superior to staff-rated scores. The best predictive model involved both the case managers' ratings from the START and the subjects' self-appraisal of risk and protective factors (area under the curve = 0.70, 95% CI 0.60–0.80).
Discussion
This review found a small but significant literature of structured approaches to including patient voice in risk assessment and management in forensic mental health care. From the early case study of Bjorkly,9 there are now three lines of research looking at structured approaches to risk assessment and management in forensic settings. These are structured violence risk intervention using the ERM, two models of staff- and patient-completed structured professional judgment tools (the DUNDRUM and START), and joint staff–patient needs assessment. While some of the data are exploratory, two studies are well-designed, controlled intervention studies.
The study showing greatest effectiveness was the ERM approach, which is notable as both a shared risk analysis and a structured intervention to reduce violence in persons with serious mental illness.8 Linking risk understanding directly to nursing interventions, Fluttert et al.8 demonstrated efficacy in reducing institutional violence. The RACE studies,12,13 which looked at the patient-rated version of the START, are impressive field tests of these concepts, although they focused on a different patient group than that found in most forensic settings. These studies demonstrated the feasibility of including client perspectives into risk and protective factors, the significant univariate association of client self-ratings to adverse outcomes, and the contribution of self-ratings to the best fit for outcomes when a multivariate analysis was used.12,13 Patient involvement in needs assessment was also found to be feasible by Abou-Sinna and Luebbers.10
Finally, the comparisons of structured patient and staff ratings of program completion and future risk needs in the DUNDRUM tools suggest that this is a promising approach to understanding patient and staff evaluation of risk and treatment progress.11 The suggestion that agreement between staff and patients on these dimensions increases later in the inpatient rehabilitation process raises the possibility that agreement between the patient and the clinician could be a useful index in progress in the recovery journey.
Taken together, these studies demonstrate the feasibility of shared approaches to risk assessment and management and present suggestive evidence of positive impact on inpatient and outpatient outcomes in certain ways. The relationships are, however, complex. Interestingly, Troquete et al.12 found no difference between the group who used the START and CSA and the control group, even though START/CSA factors were predictive of recidivism. This may mean that relevant factors identified using the START/CSA were not integrated effectively into care plans. The ERM design did include integration with interventions and showed a strong impact on institutional violence. Reduction in frequency and severity of anger incidents stresses the need of close collaboration in developing treatment strategies as well as managing risks. The validity of patient self-rating of risks is underscored by the strong correlation between patient and staff ratings of risk9,10 and the finding by van den Brink et al.13 that client ratings independently predicted violent or criminal behaviors at the six-month follow-up.
There is consensus across the studies of the significance of user involvement and collaboration in risk assessment and management. However, the paucity of research in this area is underscored by the mere handful of papers found in our review. Eidhammer et al.14 pointed out in their 2014 review of three studies of SDM in forensic mental health, two of which are included here, that there may be feasibility issues in the methods of encouraging patient collaboration by adapting tools such as the START and the DUNDRUM 3 and 4 as patient-rated tools. What works as a clinician tool may not capture patient perspectives well. Tools that require and guide staff–patient collaboration in both the risk assessment phase and the intervention phase of SDM, such as the ERM, may be more acceptable and more effective than simply having the patient complete an assessment tool. Patient collaboration needs to be a defined ongoing clinical activity assisted by tools, but it is unlikely to be achieved simply by completion of a tool as a one-time exercise. There is ample qualitative evidence for the desire for patient involvement in forensic recovery processes,6,7 and these few studies can start to guide clinicians in how to approach SDM.
Limitations and Future Research Directions
This review is limited by the small number of studies published to date, restricting this to a narrative review only. Nonetheless, we selected this approach to identify promising practices that might inform future studies. The studies are diverse and use different approaches to SDM. We believe it is important to demonstrate that SDM is both feasible and has predictive power similar to staff-rated tools. If that is so, the added benefits of patient involvement can then be explored to see if it adds incrementally to forensic outcomes.
These studies require replication and expansion. The ERM represents a model of shared clinical intervention that is very promising, but it has not been replicated. The application of the staff- and patient-rated DUNDRUM 3 and DUNDRUM 4 should be studied longitudinally to understand its potential efficacy and whether the correlation between security level and staff–patient concordance can be shown in a longitudinal design. Similarly, structured professional judgment tools with self-rated opportunities represent an integration of patients into risk-management decision-making that needs further exploration. There may be other models, such as patient involvement in shared risk-formulation parallel processes that are developing in correctional rehabilitation (e.g., Shaw et al.15). Case–control and randomized, controlled designs are needed to evaluate the effectiveness of these interventions so that the nature and magnitude of the effects of shared risk can eventually feed into evidence-based approaches to forensic care.
Most of the samples were composed of male forensic patients, and future research needs to include women. Furthermore, work needs to be done to identify the components of a successful shared risk-assessment program to facilitate practice change. Future studies should also focus on knowledge translation by identifying barriers to and facilitators of implementing these innovations in routine forensic practice.
- © 2019 American Academy of Psychiatry and the Law
